There is no defensible industry benchmark for AI visibility in dentistry, and any published average is either extrapolated from one company's client list or invented outright. What you can build instead is a local benchmark: your practice against the three competitors you actually lose to, on the twenty questions your patients ask. That takes an afternoon and tells you more than a national figure ever would.
The scoring rule lives in how to measure AI visibility for your dental practice, and the scorecard below is that same sheet read across months, not a second instrument.
Why There Is No Universal Benchmark For AI Visibility
Three properties of AI answers make a published average impossible to trust.
Personal, Local and Moving
- Personal. A signed-in session carries history. A model that assumes you already have a dentist gives a different answer to the same question, and that is not a measurement of your market.
- Local. The answer depends on which pages the engine could read for your city: a service area page, a directory entry, a review profile. Change the city and you change the evidence.
- Moving. The same question asked twice can return different names. Model updates change answers without anything on your website changing at all.
A published average fails for a second reason: the denominator. A visibility figure means nothing unless it states which questions were asked, in which city, how many times each and on which engines. A number without those four is decoration.
drsmiles.ai does not publish an industry benchmark, and we would not publish one even if our own data supported it, because a number readers cannot check is worse than no number at all. What we publish is method: how to measure, what a score actually measures, and how to compare yourself against the practices you compete with. The mechanics are in our step by step measurement guide.
The Two Numbers Worth Tracking
Most practices end up watching one number, and it is usually the wrong one.
The Number That Pays
Mention rate on money questions. Your practice named and recommended, divided by runs, across the questions where a patient is ready to book: emergency, treatment, price, insurance. This is the figure that connects to the appointment book, and it is the one worth putting in front of an owner.
Presence on open questions. The wider set: who is the best dentist around here, how do I choose a dentist, what should I ask at a first visit. Here you are rarely named. Models answer generically and list nobody, or two practices, or three. Every practice scores low on this set, which is exactly why it is useful. It is open ground. Being named where the model normally names nobody is worth more than another win inside a list everybody is already in.
Your own pages are the part of the answer set you control, and markup such as FAQ structured data makes a question and its answer explicit for a machine reading the page.
One number tells you whether you are winning a category. The other tells you whether a category is still open. A practice can improve the first substantially and stay invisible everywhere else, and the first number will still look like progress.
How To Build Your Own Local Benchmark
Three groups, one question set, one city. Run all three on the same day, inside the same few hours, with identical wording.
Building The Competitor Set
- Your practice. The same twenty questions you already use, in your own city.
- Three competitors you actually lose to. Take the names that appear in your answers when you do not, and keep those that come up most often. Our competitor map method explains how to choose them properly rather than by rivalry.
- The open category. Run the same questions with your practice name and your city removed, so the model answers from scratch. Whatever it names there is the field you are really in, and quite often the field is empty.
Then mark each of the twenty questions: you were named and recommended, a competitor was named and you were not, or nobody was named. The second column is your gap. The third is your headroom. Together they tell you where the next month goes.
Your position is a percentile inside your own city, not against the country. Say your practice is named in eleven of twenty money questions while the three local competitors are named in nine, thirteen and seven. You are second in your market. That calls for a very different plan from being sixth nationally, and it is a claim you can defend, because you produced every number in it yourself.
What Realistic Movement Looks Like
- Three to six weeks for the first movement, and it is usually small: one question flipping from absent to named after an entity fact or a directory profile was corrected.
- Two to four months for meaningful competitive movement on the money questions, if you act on the gaps you found rather than on the ones that are easiest to fix.
- Beyond that it depends on how much of the answer set you control, and that is the part nobody can promise you.
Re-running identical questions matters because change is only visible against a fixed ruler. A reworded question, a different day or a different model version gives you a number that cannot be compared with the last one. The instrument has to stay still, or the measurement moves instead of the market. Our guide to common AI search mistakes covers the habits that quietly corrupt results.
When a Score Is Misleading
- Volatile questions. Who is open right now changes legitimately, by hour and by season. Score it separately or leave it out.
- Seasonal demand. Dental questions rise with weather, school holidays and insurance enrolment windows. Compare March with March.
- Model-specific behaviour. One engine may name directories, another may name practices, and a practice can lead on one and vanish on the other.
- Personalised accounts. A session that knows your history is not a sample of your market.
- Averaging across engines. The combined figure is for reporting. Decisions come from the per-engine figures, because the engines disagree.
If your score improved and you cannot name the specific change that caused it, treat the improvement as noise until you can. Keep a written log of what you changed and when, and within two months it becomes obvious which fixes actually pay.
The Scorecard Template
One page, six lines, sixty seconds to read. Fill it in monthly and take it to whoever signs off on spend.
Swipe the table sideways to see every column.
| Line | What To Write | Where It Comes From |
|---|---|---|
| Mention rate on money questions | Named in X of Y runs | Your scoring sheet |
| Position when named | First, middle or last, and the position you usually hold | Your scoring sheet |
| Questions won | Count of money questions where you were named and recommended | Your question set |
| Questions lost | Count where a competitor was named and recommended | Your question set |
| Open categories covered | How many open questions named you at all | Your question set |
| Change since last month | Which line moved, and the change you made that caused it | Your change log |
Read the fourth line first. Questions lost tells you what to do next; the rest tells you whether it worked. If you would rather have this built for you, the full audit runs the four engine, forty question version with the scoring already wired up. The pricing and terms FAQ covers what an engagement includes, and the rest of the articles here work through the underlying mechanics in more detail.
Before you trust a monthly scorecard, check five things:
- The question set is unchanged since the first run, with no reworded prompts hiding in the file.
- Every run is recorded, including the ones that came out badly.
- Competitors are scored with identical wording, never with a prompt that names them.
- Scores are kept per engine first and combined second.
- The open category is run without your name in it, so you are not marking your own homework.